SIKs control osteocyte responses to parathyroid hormone

Parathyroid hormone (PTH) activates receptors on osteocytes to orchestrate bone formation and resorption. Here we show that PTH inhibition of SOST (sclerostin), a WNT antagonist, requires HDAC4 and HDAC5, whereas PTH stimulation of RANKL, a stimulator of bone resorption, requires CRTC2. Salt inducible kinases (SIKs) control subcellular localization of HDAC4/5 and CRTC2. PTH regulates both HDAC4/5 and CRTC2 localization via phosphorylation and inhibition of SIK2. Like PTH, new small molecule SIK inhibitors cause decreased phosphorylation and increased nuclear translocation of HDAC4/5 and CRTC2. SIK inhibition mimics many of the effects of PTH in osteocytes as assessed by RNA-seq in cultured osteocytes and following in vivo administration. Once daily treatment with the small molecule SIK inhibitor YKL-05-099 increases bone formation and bone mass. Therefore, a major arm of PTH signalling in osteocytes involves SIK inhibition, and small molecule SIK inhibitors may be applied therapeutically to mimic skeletal effects of PTH.

O steoporosis is a serious problem in our ageing population, with fragility fractures costing $25 billion annually 1 . Novel treatments are needed to boost bone mass. Osteocytes, cells buried within bone, orchestrate bone remodelling by secreting endocrine and paracrine factors 2 . Central amongst these are RANKL (encoded by the TNFSF11 gene), the major osteocyte-derived osteoclastogenic cytokine 3,4 and an FDA-approved osteoporosis drug target, and sclerostin (encoded by the SOST gene), an osteocyte-derived WNT pathway inhibitor that blocks bone formation by osteoblasts 5 and current osteoporosis drug target 6 .
When given once daily, parathyroid hormone (PTH), is the only approved osteoporosis treatment agent that stimulates new bone formation. The proximal signalling events downstream of Gsa-coupled PTH receptor signalling in bone cells are wellcharacterized 7 , but how cyclic adenosine monophosphate (cAMP) generation in osteocytes is linked to gene expression changes remains unknown. SOST and RANKL are wellestablished target genes important for the physiological effects of PTH on osteocytes. Among the mechanisms through which PTH stimulates new bone formation, down-regulation of SOST expression in osteocytes plays an important role [8][9][10] . PTH also stimulates bone catabolism, in large part through stimulation of osteoclastogenesis via inducing RANKL [11][12][13][14] , which may limit its therapeutic efficacy.
We have previously described a role for the class IIa histone deacetylase HDAC5 as a negative regulator of MEF2C-driven SOST expression, both in vitro in Ocy454 osteocytic cells 15 and in vivo 16 . Class IIa HDACs are uniquely endowed with N-terminal extensions that allow them to sense and transduce signalling information 17 . When phosphorylated, class IIa HDACs are sequestered in the cytoplasm via binding to 14-3-3 proteins. When de-phosphorylated, they are able to translocate to the nucleus to inhibit MEF2-driven gene expression 18 . Like class IIa HDACs, cAMP-regulated transcriptional coactivators (CRTC) proteins shuttle from the cytoplasm to the nucleus where they function as CREB coactivators 19 . Both HDAC4/5 and CRTC2 are known substrates of salt inducible kinases (SIKs) [19][20][21][22] , and SIK3 deficiency in growth plate chondrocytes increases nuclear HDAC4 and delays MEF2-driven chondrocyte hypertrophy 21 .
Here, we show that PTH signalling in osteocytes uses both HDAC5 and the closely related family member HDAC4 to block MEF2C-driven SOST expression. In addition, PTH-stimulated RANKL expression requires CRTC2. PTH signalling, via cAMP, inhibits SIK2 cellular activity in osteocytes. SIK inhibition, both in vitro and in vivo, achieved via the small molecule YKL-05-093, is sufficient to mimic many of the effects of PTH, including lower levels of HDAC4/5/CRTC2 phosphorylation, SOST inhibition and RANKL stimulation. Strikingly, a major arm of PTH signalling in osteocytes involves SIK inhibition, as revealed by RNA-seq analysis of PTH-versus YKL-05-093-treated osteocytes. Finally, we demonstrate that YKL-05-099 (ref. 23), an analogue of YKL-05-093 with properties making it suitable for targeting SIKs in vivo, is able to boost osteoblast numbers, bone formation, and bone mass in mice. In summary, our results demonstrate that a PTH receptor/cAMP/SIK/class IIa HDAC/CRTC axis has a crucial role in osteocyte biology.

Results
Class IIa HDACs control bone mass through SOST. Having previously demonstrated that HDAC5 blocks MEF2C-driven SOST expression in osteocytes 16 , we sought to determine whether HDAC5 and SOST interact in vivo to control bone mass. Two complementary approaches demonstrated that this was the case. First, compound heterozygosity of HDAC5 and SOST rescued the cortical and trabecular high bone mass phenotype of SOST þ / À mice ( Supplementary Fig. 1A-C). Second, anti-sclerostin antibody treatment rescued the trabecular osteopenia present in HDAC5 À / À animals ( Supplementary Fig. 1D), which have high levels of SOST expression 16 .
With evidence that HDAC5 control of SOST is physiologically important, we asked if other class IIa HDACs function in osteocytes. We 16 and others 24 have previously reported that HDAC5 À / À mice display mild trabecular osteopenia. For these studies, we extended our analyses to include the closely related family member HDAC4 for two reasons. First, endogenous MEF2C immunoprecipitates from Ocy454 cells contained HDAC4 in addition to HDAC5 (Fig. 1a and ref. 16). Second, while no obvious skeletal phenotype was observed when HDAC4 was deleted from osteocytes using DMP1-Cre 25 , compound deletion of both HDAC4 and HDAC5 led to a skeletal phenotype not observed in either single mutant strain, characterized by severe trabecular osteopenia (Supplementary Table 1 and Supplementary Fig. 1F for results of static and dynamic histomorphometry results), increased osteocyte density (Fig. 1b,c), disorganized, 'woven' cortical bone (Fig. 1d), failure to respond to sclerostin antibody ( Supplementary Fig. 1D), and reduced endocortical bone formation ( Supplementary Fig. 1E). As we previously reported, mice lacking HDAC5 alone show mild cancellous osteopenia and reduced markers of bone formation by histomorphometry 16 .
PTH signals through HDAC4 and HDAC5 to suppress SOST. We next asked whether PTH, a known suppressor of SOST expression 8 , worked through HDAC4, HDAC5, or both. PTH treatment of Ocy454 cells caused translocation from the cytosol to the nucleus of both HDAC4 and HDAC5 (Fig. 2a). When phosphorylated, class IIa HDACs are predominantly cytoplasmic through retention by 14-3-3 proteins 17 . When dephosphorylated, class IIa HDACs translocate to the nucleus where they potently inhibit MEF2-driven gene expression in muscle 26,27 . In neurons, HDAC5 nuclear import is additionally inhibited by CDK5mediated phosphorylation at S279 (ref. 28). PTH signalling reduced phosphorylation of HDAC4 at S246/S632 and, to a lesser extent, HDAC5 at S259/S279 (Fig. 2b, Supplementary Fig. 2A). Others have over-expressed HDAC5 in a rat osteosarcoma cell line to demonstrate that mutation of these serines to alanine led to PTH-independent nuclear import 29 . PTH-induced loss of phosphorylation and nuclear translocation of HDAC4/5 requires cAMP signalling, as evidenced by the fact that these events did not occur in cells lacking Gsa via CRISPR/Cas9-mediated genome editing (Fig. 2c,d, and Supplementary Fig. 2B-E). As previously described 15,30,31 , Gsa deficiency significantly increases sclerostin production by osteocytes. However, reducing MEF2C levels via shRNA or by over-expressing a constitutively nuclear super-repressor form of HDAC5 rescued this phenotype (Supplementary Fig. 2F-I), consistent with the model that Gsa deficiency increases sclerostin production via a gain-of-function MEF2C phenotype.
To determine the relevance of HDAC4/5 in mediating PTH actions in vivo, HDAC4/5-deficient mice were treated with PTH, and acute effects were measured 90 min later. While bone RANKL levels increased comparably across all four genotypes (WT, HDAC5 À / À , HDAC4f/f;DMP1-Cre, and HDAC4f/f;HDAC5 À / À ; DMP1-Cre), HDAC4/5-deficient mice were unique in that SOST levels failed to decrease following PTH treatment (Fig. 3a,b). At the protein level, PTH administration significantly decreased the numbers of sclerostin-immunoreactive cortical osteocytes in all genotypes tested except in HDAC4/5-deficient animals (Fig. 3c,d). Taken together, these results indicate that HDAC4 and HDAC5 are downstream of PTH receptor signalling, and are required for PTH-mediated SOST suppression, both in vitro and in vivo.
While SOST is a well-studied PTH target genes, it represents a small portion of the transcriptome regulated by parathyroid hormone (see below). Underscoring this point, once daily intermittent PTH treatment leads to comparable gains in trabecular bone density in mice lacking HDAC4 in osteocytes, HDAC5 or both ( Supplementary Fig. 4A). Therefore, although class II HDACs are required for acute PTH-induced changes in SOST expression, other signalling arms and target genes downstream of the PTH receptor must exist that are important for the pharmacologic effects of this hormone.
SIK2 is inhibited by PTH and required for PTH signalling. We next addressed the signalling mechanisms used between activation of the PTH receptor and decreased phosphorylation of HDAC4/5. In chondrocytes in vitro, PTHrP drives HDAC4 into the nucleus via PP2A-mediated dephosphorylation, which can be blocked by okadaic acid 34 . Surprisingly, okadaic acid did not block PTH-mediated decreased HDAC4/5 phosphorylation or SOST suppression in Ocy454 cells ( Supplementary Fig. 4B,C). Similarly, PTH-induced decreases in HDAC4/5 phosphorylation and SOST suppression were intact when PP2A catalytic subunit levels were reduced via shRNA ( Supplementary Fig. 4D    Okadaic acid and PP2A shRNA efficacy was confirmed in these experiments based on observed increases in HDAC4 S246 phosphorylation ( Supplementary Fig. 4B,D). Taken together, these results suggest that, unlike in chondrocytes, in osteocytes PTH-stimulated decreased phosphorylation of HDAC4/5 is not mediated by activation of PP2A. To explore candidate kinases whose activity might mediate the actions of PTH on HDAC4/5, we examined salt inducible kinases (SIKs), AMPK family members reported to function as class IIa HDAC N-terminal kinases 20,35 . Subcellular fractionation experiments revealed that both SIK2 and SIK3 proteins are predominantly cytoplasmic in osteocytes (Supplementary Figure 4F). Combined silencing of both SIK2 and SIK3 in Ocy454 cells significantly decreased HDAC4/5 N-terminal phosphorylation (Fig. 4a).
cAMP signalling in adipocytes and hepatocytes inhibits SIK2 activity via protein kinase A (PKA)-mediated phosphorylation, which in turn sequesters SIK2 from its substrates [36][37][38] . PTH signalling in osteocytes triggered SIK2 phosphorylation at S343, S358 and T484 (Fig. 4b). PKA-mediated SIK3 phosphorylation was not triggered by PTH signalling (Fig. 4b). Notably, PTH-stimulated SIK2 S358 phosphorylation occurred rapidly, faster than the fall in HDAC4/5 phosphorylation levels (Fig. 4c). Importantly, SIK2-silenced cells showed normal up-regulation of the PTH target gene CITED1 (ref. 39) (Fig. 4d). In contrast, PTHinduced decreases in HDAC4/5 phosphorylation (Fig. 4e) and SOST suppression (Fig. 4f) did not occur in SIK2-silenced cells. Interestingly, PTH-induced RANKL upregulation, an HDAC4/5independent phenomenon (Figs 2f and 3a) also did not occur in SIK2-deficient osteocytes (Fig. 4g), suggesting that another SIK substrate is involved in PTH-mediated RANKL gene induction. SIK3-deficient cells showed normal PTH responses (Fig. 4d,f,g), as predicted by the fact that this protein is not phosphorylated in response to PTH signalling. cAMP responses to PTH were blunted in SIK2-silenced Ocy454 cells but were clearly present at PTH levels above 4 nM ( Supplementary Fig. 4G). Nevertheless, this effect on cAMP levels in response to PTH is unlikely to explain the effects of SIK2 silencing. Forskolin-induced cAMP up-regulation was normal in SIK2-deficient cells, yet this agent failed to regulate SOST or RANKL expression in the absence of SIK2 (Fig. 4h).
To determine the relevance of SIK2 in mediating PTH actions in vivo, mice lacking SIK2 in DMP1-expressing cells (including osteocytes) were treated with PTH, and acute effects were measured in bone 120 min later. Figure 4i shows that DMP1-Cre deletion of SIK2 led to a significant reduction in SIK2, but not PTH receptor, mRNA levels in bone. Similar to the results in Ocy454 cells, PTH-induced CITED1 up-regulation was preserved in SIK2 OcyKO mice (Fig. 4j). However, PTH-induced SOST and RANKL gene regulation did not occur in the absence of SIK2 (Fig. 4k).
RANKL is a known PTH target gene; previous studies have suggested an important role for CREB, through binding to an enhancer 75 kB upstream of the transcription start site [12][13][14]40,41 . While CREB itself is not a known SIK substrate, the CRTC CREB coactivator proteins are 19 . All three CRTC proteins are expressed in osteocytes; therefore, levels of each were reduced individually using shRNA. Only CRTC2 silencing was sufficient to block PTH-induced RANKL up-regulation (Fig. 4l). Supplementary Fig. 4J shows that PTH-induced cAMP generation was normal in CRTC2-deficient cells. PTH promoted CRTC2 nuclear translocation in a Gsa-dependent manner (Fig. 2c), and CRTC2 inducibly associated with the À 75 kB 'D5' RANKL enhancer 42 following PTH treatment (Fig. 4m). In summary, these results demonstrate that two key SIK substrates, HDAC4/5 and CRTC2, play major roles in PTH-mediated regulation of SOST and RANKL expression, respectively. SIK inhibitors regulate SOST and RANKL expression. Gene ablation studies in vitro and in vivo suggested that SIK2 is required for PTH to regulate SOST and RANKL expression, and that PTH signalling leads to PKA-mediated SIK2 inhibition. Therefore, we wondered whether acute inhibition of SIK kinase activity in otherwise normal cells or mice would be sufficient to mimic these actions of PTH. HG-9-91-01 is a small molecule kinase inhibitor with demonstrated biologic activity against SIKs in cultured macrophages, dendritic cells and hepatocytes 37,38,43,44 . However, HG-9-91-01 is not SIK-specific and is not suitable for in vivo use; therefore, we screened for analogues based on the goals of improved specificity and pharmacokinetics. These efforts ultimately led to the identification of YKL-04-114 and its closely related analogue YKL-05-093 (Fig. 5a). The K d of YKL-05-093 for SIK2 is 7.1 nM, and its activity against a panel of 96 recombinant kinases is shown in Supplementary Table 2 (here SIK refers to SIK1 and QSK refers to SIK3) and shown graphically in Supplementary  Fig. 5A. YKL-04-114 or YKL-05-093 treatment of Ocy454 cells led to rapid, dose-dependent decreases in HDAC4/5 phosphorylation (Fig. 5b,c), and increased nuclear translocation of HDAC4 and CRTC2 (Fig. 5d). YKL-04-114 caused rapid and potent SOST suppression and RANKL up-regulation ( Importantly, treatment with YKL-05-093 did not decrease HDAC4 S246 phosphorylation or cause SOST suppression in osteocytes lacking SIK2 and SIK3 ( Fig. 5f,g). In addition, PTH and YKL-05-093-mediated stimulation of RANKL expression was abrogated in cells lacking CRTC2 (Fig. 5h). So although YKL-05-093 does target other kinases in vitro, its cellular actions studied here depend on the presence of SIK2 and SIK3.
On the basis of our model that YKL-05-093 functions as a SIK inhibitor downstream of PTH-stimulated cAMP generation, one would predict that the inhibitor would regulate gene expression in Gsa-deficient osteocytes. Indeed, YKL-05-093 treatment of Gsa-deficient Ocy454 caused SOST suppression and RANKL up-regulation with effects similar to forskolin, except, as expected based on its inability to increase cellular cAMP levels ( Supplementary Fig. 5B), YKL-05-093 did not increase SIK2 S358 phosphorylation (Fig. 5i,j).
Small molecule SIK inhibitors mimic PTH action in vitro. The ability of YKL-05-093 to mimic the effects of PTH with respect to SOST and RANKL gene regulation supports the hypothesis that the actions of YKL-05-093 might mimic the effects of PTH on many genes. We therefore performed RNA-seq on Ocy454 cells treated for four hours with vehicle, PTH (1 nM) or YKL-05-093 (0.5 mM) to determine the overlap in global gene regulation by these two agents. Significantly 446 genes were (42 fold, FDRo0.05) regulated by PTH, and 257 genes were significantly regulated by YKL-05-093. Of the 446 PTH-regulated genes, 142 (32%) were co-regulated in the same direction by YKL-05-093 ( Fig. 6a,b, Supplementary Table 3 for differentially expressed genes and Supplementary Dataset 1 for all RNA-seq data). This significant overlap was not due to random chance ( Supplementary Fig. 6A,B). Gene ontology analysis for the genes regulated by both PTH and YKL-05-093, is shown in Supplementary Fig. S6C,D: many of the co-regulated genes fit into categories of interest such as 'ossification' and 'mesenchyme development'.
YKL-05-093 mimics PTH actions in vivo. While YKL-04-114 and YKL-05-093 had comparable activity in vitro, YKL-05-093 showed improved stability when exposed to murine hepatic microsomes in vitro ( Supplementary Fig. 7). Therefore, mice were treated with YKL-05-093 and effects on gene expression in bone were assessed 2 h later. Similar to acute PTH administration (Fig. 3), intraperitoneal YKL-05-093 administration led to dose-dependent SOST suppression and RANKL up-regulation in osteocyte-enriched bone RNA (Fig. 7a,b). This was accompanied by reductions in sclerostin protein levels measured by immunohistochemistry (Fig. 7c). Finally, expression of genes identified by RNA-Seq as co-regulated by PTH and YKL-05-093 in vitro were measured: as shown in Fig. 7d-i, in vivo 20 umol kg À 1 YKL-05-093 treatment leads to significant regulation of VDR, WNT4, NR4A2, NUAK1, PDGFA and CD200 expression in the directions predicted from the in vitro experiments. Therefore, acute YKL-05-093 treatment in vitro and in vivo engages a program of gene expression quite similar to one used by parathyroid hormone, thus identifying SIK inhibition as an important mechanism used by PTH to regulate gene expression in osteocytes.
Male mice were then treated with vehicle or YKL-05-099 (6 mg kg À 1 ) once daily via intraperitoneal injection for 2 weeks. Bone RNA from these animals revealed that RANKL levels were increased and there was a trend towards reduced SOST (Fig. 8d). In addition, genes expressed by osteoblasts (osteocalcin (encoded by the BGLAP gene) and COL1A1) were significantly increased by YKL-05-099 treatment, suggesting possible positive effects on osteoblastic bone anabolism (Fig. 8d). To determine effects on bone mass and cellular composition/activity, static and dynamic histomorphometry were performed. Indeed, once daily YKL-05-099 treatment increased cancellous bone mass (Fig. 8e) and osteoid surface (Fig. 8f), suggesting accelerated bone formation. Dynamic histomorphometry revealed that YKL-05-099 led to increased mineralizing surface, a trend towards increased matrix apposition rate, and increased bone formation rate (Fig. 8g,h,i,l). At the cellular level, YKL-05-099 treatment increased osteoblast numbers (Fig. 8j,m) and reduced osteoclast numbers (Fig. 8k). Other than the observed reduction in osteoclast numbers (see Discussion), these findings are quite similar to the effects of once daily PTH treatment.

Discussion
PTH is currently the only approved osteoporosis therapy that promotes new bone formation. While its effects on target cells in bone are broad, major target genes in osteocytes responsible for its ability to increase both bone formation and resorption include SOST and RANKL, respectively. Here we demonstrate that SIKs act as gatekeepers to regulate a major arm of PTH signalling in osteocytes, including (but not limited to) these two important target genes. Tonic SIK activity leads to constitutive phosphorylation and cytoplasmic localization of HDAC4/5 and CRTC2. Activation of protein kinase A, as occurs with activation of the PTH receptor 7 , leads to multisite phosphorylation on SIK2, modifications that inhibit its cellular activity 36,38 . This inhibition reduces tonic HDAC4/5 and CRTC2 phosphorylation, which in turn leads to their nuclear localization and action on respective target genes (Fig. 9). Whether this pathway operates in other PTH/PTHrP target cells, such as chondrocytes 45 , renal epithelial cells 46 , T lymphocytes 47 and adipocytes 48 remains to be determined. HDAC4/5 are required for PTH-stimulated SOST repression in osteocytes, through effects on MEF2C binding to the þ 45 kB SOST enhancer. Previous overexpression studies have suggested that PTH signalling impinges on the upstream SOST enhancer 29,49,50 : here we show that HDAC4/5 are required for this effect using loss of function approaches in vitro and in vivo. At later time points, PTH treatment reduces in MEF2C mRNA levels 32,33,51 , in addition to the post-translational effects on DNA binding observed here earlier (Fig. 2g). Similarly, PTH induces both the rapid nuclear translocation of HDAC4 and, at later time points, increases in HDAC4 mRNA ( Fig. 6 and   sclerostin 55 (Supplementary Fig. 1). Notably, sclerostin transgenic mice do not display woven bone and increased osteocyte density, and sclerostin antibody did not increase BMD in DKO animals. Therefore, class IIa HDACs control expression of additional genes in osteocytes that potently regulate skeletal biology. In addition, as evident by the fact that HDAC4/5 'DKO' mice show a preserved bone anabolic effect of intermittent PTH treatment ( Supplementary Fig. 4A), class II HDAC/SOST-independent pathways that mediate the pharmacologic effects of parathyroid hormone must exist. Interesting parallels and distinct differences are noted between PTH-mediated SOST suppression in osteocytes and PTHrPmediated suppression of expression of the Collagen X gene in growth plate chondrocytes 34 . While both pathways utilize a class IIa HDAC/MEF2 mechanism of action, the signalling events required for HDAC4 nuclear translocation may differ. PTH signalling in osteocytes involves inhibition of SIK activity, while in chondrocytes, PTHrP signalling activates the cAMP-dependent phosphatase PP2A. That being said, a role for SIKs in PTHrP signalling in chondrocytes cannot be excluded given the fact that SIK3-deficiency 21 appears to phenocopy the effects of PTHrP overexpression 56 . Our experiments with okadaic acid and PP2A shRNA ( Supplementary Fig. 4A-D) argue against a major role for PP2A in mediating PTH signalling in osteocytes. Because the inhibition of HDAC4/5 phosphorylation in response to PTH was substantial, any further action of PTH on PP2A or other phosphatases would be likely to have a modest effect on overall phosphorylation levels.
PTH signalling to regulate RANKL expression in osteoblasts and osteocytes has been studied extensively over the past decade. Many investigators have demonstrated a role for a cAMP/CREB pathway via the gene's upstream enhancers 13,40,42,57 . Here we show an additional requirement for the presence of a CREB co-activator, CRTC2, for PTH-induced RANKL gene regulation. It is of interest that PTH action requires two pathways, one involving a direct PKA target (CREB) and another that uses PKA-mediated SIK inhibition. Since SIK inhibition, through suppression of SOST expression, can also increase bone formation, one can speculate that this use of the SIK pathway 'forces' PTH action to link bone resorption and bone formation.
A recent report has suggested that, in osteoblasts, PTH signalling promotes proteasomal degradation of HDAC4 that in turn allows MEF2C-driven activation of the RANKL promoter 24 . We do not observe changes in HDAC4/5 levels after PTH treatment, which may be explained by the differing time courses and cell types used. We favour a model in which PTH induces RANKL expression in osteocytes via its À 75 kB enhancer through SIK-dependent CRTC2 nuclear translocation.
The use of SIK inhibitors uniquely allows us to examine the acute effects of changes in SIK enzyme activity in cells and mice. These experiments show that the effects of SIK inhibition are rapid enough to mediate the effects of PTH on SOST and RANKL expression. In this way, though the inhibitors are less specific than gene knockout or shRNA-mediated expression knockdown, their use complements the data derived from the genetic studies. While YKL-05-093 and YKL-05-099 do inhibit kinases other than SIKs when tested in vitro, many of their effects in Ocy454 cells, including those on SOST and RANKL expression, were not observed when SIK2/3 proteins were absent.
The role of SIK2 and SIK3 (the predominant SIKs expressed in osteocytes) in bone biology in vivo remains incompletely understood. Global SIK2 knockout mice have been shown to display phenotypes in melanocytes 58 , neurons after ischaemic injury 59 , cardiomyocytes during hypertrophy 60  In the absence of PTH signaling, SIK2 tonically phosphorylates its substrates HDAC4/5 and CRTC2, leading to their cytoplasmic retention via binding to 14-3-3 chaperones. PTH signaling leads to PKA-mediated phosphorylation of SIK2, which inhibits its cellular activity. This in turn reduces phosphorylation of HDAC4/5 and CRTC2, leading to their dephosphorylation by an unknown phosphatase (ppase), and subsequent nuclear translocation. Small molecule SIK inhibitors (YKL-05-093 and YKL-05-099) mimic the effects of PTH by directly blocking SIK2 kinase activity. In the nucleus, HDAC4/5 block MEF2C-driven SOST expression, while CRTC2 enhances CREB-mediated RANKL gene transcription. PTH-induced reductions in sclerostin contribute to increased bone formation, while PTH-induced increases in RANKL drive increased bone resorption.
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms13176 ARTICLE homoeostasis 61 . Conditional SIK2 mutant alleles have been described 38,62 to further study the role of this kinase in hepatocytes and in the pancreas. Global SIK2 knockout mice have no skeletal phenotype reported to date. Here, we have deleted SIK2 from DMP1-Cre expressing cells, and have observed that this gene is required for the acute response of osteocytes to PTH. A detailed description of the global bone phenotype of the SIK2 OcyKO strain remains to be determined. Global SIK3 deficient mice display a dramatic growth plate phenotype 21 that confounds study of direct actions of SIK3 in osteocyte biology in vivo.
A conditional SIK3 allele has been reported, and deletion in chondrocytes confirms the cell-intrinsic role for SIK3 in these cells 63 . No studies to date have examined the role of SIK3 in osteocytes in vivo. SIK inhibition downstream of cAMP signalling has long been appreciated to occur 36 , but the relative contribution of SIK inhibition to overall changes in gene expression due to Gsa-coupled GPCR signalling has not previously been explored. Remarkably, 32% of genes regulated by PTH in osteocytes were co-regulated by YKL-05-093. While it is likely that many of these genes (like SOST and RANKL) are regulated in turn by HDAC4/5 and CRTC2, undoubtedly additional SIK2/3 substrates may be responsible for these widespread effects.
Recently, pterosin B was reported as a small molecule inhibitor of SIK3 with in vivo activity in a SIK3-dependent murine osteoarthritis model 63 . Interestingly, this small molecule leads to ubiquitin-dependent SIK3 degradation, and therefore acts in a manner distinct to that of YKL-05-093 and YKL-05-099, which function as kinase inhibitors 23 . While SIK2 deficiency was sufficient to abrogate responses to parathyroid hormone in vitro and in vivo (Fig. 4), combined SIK2 and SIK3 deficiency was required to blunt effects of YKL-05-093 and YKL-05-099. This is consistent with potential redundancy between these two kinases 38 , and the fact that both inhibitors potently target SIK3 in addition to SIK2.
In many regards, YKL-05-099 treatment mimics the effects of once-daily PTH treatment in vivo. However, one notable exception is present. PTH treatment increases osteoclastic bone resorption, in part due to PTH-induced RANKL up-regulation 12 . Although YKL-05-099 potently increases RANKL levels in bone (Fig. 8d), osteoclast numbers are actually decreased by this treatment (Fig. 8j). In addition to targeting SIK2, YKL-05-099 inhibits the tyrosine kinases Src and CSF-1R (ref. 23). Src deficiency leads to functional osteoclast defects and osteopetrosis 64 , and CSF-1R deficiency causes osteoclast-poor osteopetrosis 65 . Therefore, combined SIK and Src/CSF-1R inhibition may lead to the desirable therapeutic combination of increased bone formation and reduced bone resorption. More detailed assessment of the long-term safety profile of YKL-05-099 will be required to determine if its profile of kinase inhibition will be well-tolerated over time.
Recombinant PTH is the only current osteoanabolic therapy approved for osteoporosis treatment. Our data highlight that distinct signalling modules exist downstream of PTH receptor signalling, including a major arm involving SIK inhibition. SIK inhibition is sufficient to reduce sclerostin levels and to mimic many of the other effects of PTH in osteocytes at the level of gene expression. Furthermore, in vivo SIK inhibition with YKL-05-099 boosts osteoblast numbers, osteoblast activity and trabecular bone mass. Specific inhibitors of SIK action might provide a novel approach to mimic PTH action to stimulate bone anabolism.

Methods
Animal studies. All animals were housed in the Center for Comparative Medicine at the Massachusetts General Hospital, and all experiments were approved by the hospital's Subcommittee on Research Animal Care. HDAC5-null mice 66 and HDAC4 f/f mice 67 were generously provided by Dr Eric Olson (University of Texas Southwestern Medical Center, Dallas, TX) and were backcrossed to C57B/6 mice for at least 6 generations. DMP1-Cre mice 25 were generously provided by Dr Jian (Jerry) Feng (Texas A&M University, Baylor College of Dentistry, Dallas, TX). 'DKO' HDAC4/5 mice were of the following genotype: HDAC4f/f;HDAC5 À / À ;DMP1-Cre. SIK2 f/f mice were as described 38 , and were bred to DMP1-Cre animals to generate SIK2 OcyKO mice. ES cells carrying the targeted SOST allele Sost tm1(KOMP)Vlcg , in which the SOST coding sequence has been replaced by LacZ and floxed Neo cassette, were obtained from the knockout mouse project (KOMP) repository. Clone VG10069-BE8 was injected into blastocysts, and the resulting SOST þ / À mice were crossed to HDAC5 mutant animals to generate compound heterozygous mice. In all instances, skeletal phenotypes were evaluated in 8 weekold sex-matched littermates. For acute effects of PTH on bone gene expression, animals were treated with PTH (1-34, 300 mg kg À 1 , subcutaneous administration) and then killed 90 min later. For acute effects of YKL-05-093 on bone gene expression, animals were treated with the indicated doses of compound (dissolved in PBS þ 25 mM HCl) or solvent via intraperitoneal injections and killed 2 h later. Experiments with YKL-05-099 were performed in a similar fashion: compound was dissolved in PBS þ 25 mM HCl and injected IP once daily five times per week for a total of 10 injections. For in vivo sclerostin antibody treatment, mice were treated twice weekly with sclerostin antibody (50 mg kg À 1 , subcutaneous administration, generously provided by Dr Michael Ominsky, Amgen) for 6 weeks. Power calculations were performed based on pilot experiments in which s.d.s and magnitudes of effect sizes were estimated. For experiments in which mice were treated with either vehicle or PTH (or YKL-05-093), mice were assigned to alternating treatment groups in consecutive order. Cell culture. For all experiments, a single cell subclone of Ocy454 cells 15,16 was used. Cells were passages in alpha-MEM supplemented with 10% heat-inactivated fetal bovine serum and 1% antibiotics (penicillin/streptomycin, Fungizone) at 33°C with 5% CO 2 . Cells were plated at 50,000 cells ml À 1 and allowed to reach confluency at 33°C (typically in 2-3 days). At this point, cells were transferred to 37°C for subsequent analysis. For immunoblotting, cells were always analysed after culture at 37°C for 7 days. For gene expression analysis, cells were analysed after culture at 37°C for 14 days. Mycoplasma contamination was ruled out by PCR. Cells were routinely assayed for SOST expression at 37°C and examined for osteocytic morphology.
shRNA infections and CRISPR/Cas9-mediated gene deletion. See Supplementary Table 4 for all shRNA and sgRNA targeting sequences used. For shRNA, lentiviruses were produced in 293T cells in a pLKO.1-puro (Addgene, plasmid 8453) backbone. Viral packaging was performed in 293T cells using standard protocols (http://www.broadinstitute.org/rnai/public/resources/protocols). Briefly, 293T cells were plated at 2.2 Â 10 5 ml À 1 and transfected the following day with shRNA-expressing plasmid along with psPAX2 (Addgene plasmid 12260) and MD2.G (Addgene plasmid 12259) using Fugene-HD. Medium was changed the next day, and collected 48 h later. For experiments with SIK2/SIK3 double knockdown, one shRNA was transferred into a blasticidin resistance-conferring backbone (Addgene, plasmid 26655). Cells were exposed to lentiviral particles (MOI ¼ 1) overnight at 33°C in the presence of polybrene (5 mg ml À 1 ). Media was then changed and puromycin (2 mg ml À 1 ) and/or blasticidin (4 mg ml À 1 ) was added. Cells were maintained in selection medium throughout the duration of the experiment. HDAC5 S/A complementary DNA (cDNA) was introduced via lentivirus as described in ref. 16. Briefly, control and Gsaknockdown cells were infected with lentiviral particles expressing GFP and/or HDAC5 S249/498A. After 24 h, cells were selected with hygromycin (100 mg ml À 1 ) and used for subsequent experiments.
For sgRNA experiments, first Ocy454 cells were stably transduced with a hygromycin resistance-conferring Cas9-expressing lentivirus to ensure no effects on sclerostin secretion. Sclerostin ELISAs were performed exactly as described in ref. 16. For subsequent experiments, sgRNA sequences were subcloned into PX458 (a gift from Dr Feng Zhang, Addgene plasmid 48138 (ref. 68)), a plasmid that co-expressed sgRNA, Cas9 and eGFP. Ocy454 cells were transfected with this plasmid using Fugene HD (Promega, Madison, WI) (1 mg plasmid per well of a six well plate). 48 h later, eGFP hi cells were recovered by FACS-based sorting and plated in 96 well plates at 1 cell per well. Media was changed once weekly, and 3 weeks later colonies were identified by visual inspection. Colonies were then expanded and analysed for loss of target protein expression by immunoblotting. For HDAC4 and Gs,a targeting experiments, at least three independent clones (deriving from two independent sgRNA sequences) were analysed and showed similar results. Allele-specific sequencing of mutant clones was performed by amplifying the genomic region of interest surrounding the targeted site by PCR. PCR products were then TOPO-TA cloned (ThermoFisher), and multiple bacterial colonies sequenced using T7 sequencing primer.
Real-time quantitative PCR. Total RNA was extracted from cultured cells using RNeasy (Qiagen, Venlo, Netherlands) following the manufacturer's instructions. For long bone RNA isolation, mice were killed and both femurs were rapidly dissected on ice. Soft tissue was removed and epiphyses cut. Bone marrow cells were then removed by serial flushing with ice-cold PBS. TRIzol (Life Technologies) was added and sampled were frozen at À 80°C and then homogenized. RNA was then extracted per the manufacturer's instructions, and further purified on RNeasy microcolumns before cDNA synthesis. RNA with a A260/280 ratio o1.7 was not used for downstream analysis. For cDNA synthesis, 1 mg RNA was used in synthesis reactions according to the instructions of the manufacturer (Primescript RT, Takara). SYBR Green-based quantitative PCR (qPCR) detection was performed using FastStart Universal SYBR Green (Roche, Basel, Switzerland) on a StepOne Plus (Applied Biosystems, Carlsbad, CA) thermocycler. All PCR primer sequences are listed in Supplementary Table 4.
Immunoprecipitation and immunoblotting. Whole cell lysates were prepared using TNT buffer (20 mM Tris-HCl pH 8, 200 mM NaCl, 0.5% Triton X-100 supplemented with 1 mM DTT, 1 mM NaF and protease inhibitors (Pierce, catalogue #88266). This lysis buffer was used for all experiments except those in which SIK2 and SIK3 phosphorylation was measured using phospho-specific antibodies: for those experiments, cells were lysed in buffer containing 50 mM Tris-HCl pH 7.5, 270 mM sucrose, 1 mM EDTA, 1 mM EGTA, 1 mM NaF, 1 mM DTT and protease inhibitors (Sigma, P8340). MEF2C (ref. 16) and SIK3 (refs 37,38) immunoprecipitations were performed as described. Briefly, lysates were precleared with protein A/G, then 0.5 mg total protein incubated with 1 mg antibody overnight at 4°C. The next morning, immune complexes were precipitated with protein A/G agarose, washed three times in ice-cold lysis buffer, and precipitated proteins eluted by boiling in SDS-sample buffer at 95°C for 5 min. Subcellular fractionation was performed using a commercially-available kit (Thermo Scientific, product number 78840) following the manufacturer's instructions. Lysates (15-20 mg cellular protein) were separated by SDS-polyacrylamide gel electrophoresis (SDS-PAGE), and proteins were transferred to nitrocellulose. Membranes were blocked with 5% milk in TBST, and incubated with primary antibody overnight at 4°C. The next day, membranes were washed, incubated with appropriate horseradish peroxidase (HRP)-coupled secondary antibodies, and signals detected with enhanced chemiluminescence (ECL, Pierce). All immunoblots were repeated at least twice with comparable results obtained. Supplementary Figure 8 shows the full blot corresponding to the scanned portions shown in the main text figures.
Histology and immunohistochemistry. Formalin-fixed paraffin-embedded decalcified tibia sections from 8 week-old mice were obtained. Sirius red staining was performed using Sirius red and picric acid obtained from Sigma. Sections were visualized under polarized light. Hematoxylin and eosin (H þ E) staining was performed on some sections using standard protocols, and osteocyte density was assessed on cortical bone osteocytes in a medium power field 3 mm below the tibial growth plate. For anti-sclerostin immunohistochemistry, antigen retrieval was performed using proteinase K (20 mg ml À 1 ) for 15 min. Endogenous peroxidases were quenched, and slides were blocked in TNB buffer (Perkin-Elmer), then stained with anti-sclerostin antibody at a concentration of 1:200 for 1 h at room temperature. Sections were washed, incubated with HRP-coupled secondary antibodies, signals amplified using tyramide signal amplification and HRP detection was performed using 3,3 0 -diaminobenzidine (DAB, Vector) for 2-3 min. Slides were briefly counterstained with hematoxylin before mounting. Quantification of sclerostin positive osteocytes was performed on a blinded basis. All photomicrographs were taken 3 mm below the growth plate on the lateral side of the tibia. All osteocytes were counted and then scored as either sclerostinpositive or negative. Sections from at least four mice per experimental group were analysed. Quantification of immunostaining was done based on coded sample numbers in a completely blinded manner. Representative photomicrographs are displayed next to quantification in data figures.
Chromatin immunoprecipitations. ChIP assay was performed using a kit (EZ-Chip, Miilipore, 17-371, Billerica, MA) according to the manufacturer's instructions. Briefly, cells were grown at 37°C for 7 days, followed by PTH treatment (25-50 nM) for the indicated times. Cells were then cross-linked with 1% formaldehyde for 10 min and then quenched with 0.125M glycine. Cells were lysed and sonicated with 10 pulses for 30 s each to fragment DNA to 200-800 bp fragments. DNA-protein complexes were precipitated using 1.5 mg antibodies (MEF2C, CRTC2 or control rabbit IgG) overnight at 4°C. Immune complexes were precipitated, DNA was purified and real-time PCR was conducted using primer sets (Supplementary Table 4) to detect the þ 45 kB SOST enhancer and upstream RANKL enhancers. Data are expressed as relative enrichment for each antibody (above control IgG) for each primer set. Data shown represent triplicate biological repeats within experiments, and each experiment was performed at least twice.
cAMP radioimmunoassay. Cells, in 96 well plates, were treated with indicated ligands for 20 min at room temperature in the presence of the phosphodiesterase inhibitor 3-isobutyl-1-methylxanthine (IBMX, Sigma I5879, 2 mM). The medium was then removed and cells were lysed in 50 mM HCl and transferred to À 80°C. Thawed lysates were diluted 1:5 with dH 2 O, and an 10 ml aliquot was assessed for cAMP content by radioimmunoassay using 125 I-cAMP analogue as a tracer and unlabelled cAMP to generate a standard curve.
RNA-sequencing. Total RNA was subjected to ribosomal RNA (rRNA) depletion using RiboZero kit (Illumina) followed by NGS library construction using NEBNext Ultra Directional RNA Library Prep Kit for Illumina (New England Biolabs). Experimental duplicates were performed for each condition. Sequencing was performed on Illumina HiSeq 2500 instrument, resulting in an average of 33 million pairs of 50 bp reads per sample. Sequencing reads were mapped to the mouse reference genome (mm10/GRCm38) using STAR (http://bioinformatics. oxfordjournals.org/content/early/2012/10/25/bioinformatics.bts635). Gene expression counts were calculated using HTSeq v.0.6.0 (http://www.huber.embl.de/ users/anders/HTSeq/doc/overview.html) based on a current Ensembl annotation file for mm10/GRCm38 (release 75). Differential expression analysis was performed using EgdeR package based on the criteria of more than two-fold change in expression value versus control and false discovery rates (FDR) o0.05. Venn diagrams from gene set analysis were generated using genes with 41.5 fold change in expression values and FDR o0.05. Significance testing for gene set overlap was performed according to a standard hypergeometric distribution, P-values o2.2 Â 10 À 16 . The RNA-seq data are deposited in GEO under accession code GSE76932.
Small molecule synthesis. Details regarding synthesis of YKL-04-114 and YKL-05-093 are found in Supplementary Methods. Supplementary Fig. 9 shows nuclear magnetic resonance (NMR) spectra for these compounds and related intermediates.
In vitro kinase assays. ScanEDGE kinase assays panelling specificity across a panel of 96 representative kinases were performed by DiscoverX (Fremont, CA). For most assays, kinase-tagged T7 phage strains were grown in parallel in 24-well blocks in an E. coli host derived from the BL21 strain. Bacteria were grown to log phase and infected with T7 phage from frozen stock (MOI ¼ 0.4) and incubated with shaking at 32°C until lysis (90-150 min). The lysates were centrifuged (6,000g) and filtered (0.2 mm) to remove cell debris. The remaining kinases were produced in HEK-293 cells and subsequently tagged with DNA for qPCR detection. Streptavidin-coated magnetic beads were treated with biotinylated small molecule ligands for 30 min at room temperature to generate affinity resins for kinase assays. The liganded beads were blocked with excess biotin and washing with blocking buffer (SeaBlock (Pierce), 1% BSA, 0.05% Tween 20, 1 mM DTT) to remove unbound ligand and to reduce non-specific phage binding. Binding reactions were assembled by combining kinases, liganded affinity beads, and test compounds in 1 Â binding buffer (20% SeaBlock, 0.17 Â PBS, 0.05% Tween 20, 6 mM DTT). Test compounds were prepared as 40 Â stocks in 100% DMSO and directly diluted into the assay. All reactions were performed in polypropylene 384 well plates in a final volume of 40 ml. The assay plates were incubated at room temperature with shaking for 1 h and the affinity beads were washed with wash buffer (1 Â PBS, 0.05% Tween 20). The beads were then resuspended in elution buffer (1 Â PBS, 0.05% Tween 20, 0.5 mM of the non-biotinylated affinity ligand) and incubated at room temperature with shaking for 30 min. The kinase concentration in the eluates was measured by qPCR. YKL-05-093 was screened in this assay at 71 nM (ten times its K d for SIK2), and results are reported as '% control', where lower numbers indicate stronger hits.
Micro-CT. Assessment of bone morphology and microarchitecture was performed with high-resolution micro-computed tomography (mCT40; Scanco Medical, Brüttisellen, Switzerland). In brief, the distal femoral metaphysis and mid-diaphysis were scanned using 70 kVp peak X-ray tube potential, 113 mAs X-ray tube current, 200 ms integration time, and 10-mm isotropic voxel size. Cancellous bone was assessed in the distal metaphysis and cortical bone was assessed in the mid-diaphysis. The femoral metaphysis region began 1,700 mm proximal to the distal growth plate and extended 1,500 mm distally. Cancellous bone was separated from cortical bone with a semiautomated contouring program. For the cancellous bone region we assessed bone volume fraction (BV/TV, %), trabecular thickness (Tb.Th, mm), trabecular separation (Tb.Sp, mm), trabecular number (Tb.N, 1 mm À 1 ), connectivity density (Conn.D, 1 mm À 3 ), and structure model index. Transverse CT slices were also acquired in a 500 mm long region at the femoral mid-diaphysis to assess total cross-sectional area, cortical bone area, and medullary area (Tt.Ar, Ct.Ar and Ma.Ar, respectively, all mm 2 ); bone area fraction (Ct.Ar/Tt.Ar, %), cortical thickness (Ct.Th, mm), porosity (Ct.Po, %) and minimum (I min , mm 4 ), maximum (I max , mm 4 ) and polar (J, mm 4 ) moments of inertia. Bone was segmented from soft tissue using the same threshold, 300 mg HA cm À 3 for trabecular and 733 mg HA cm À 3 for cortical bone. Scanning and analyses adhered to the guidelines for the use of micro-CT for the assessment of bone architecture in rodents 69 . For the primary spongiosa region (where intermittent PTH treatment has its predominant effect) analysed in Supplementary  Fig. 4A, coronal CT slices were evaluated in a 500 mm (50 slices) region located centrally in the bone. The region of interest began 1000 mm superior to the epiphysis and included all primary spongiosa and the medullary cavity. The primary spongiosa bone region was identified by semi-manually contouring the region of interest. Images were thresholded using an adaptive-iterative algorithm. The average adaptive-iterative threshold of control mice (WT, vehicle treated) for the region of interest (299 mgHA cm À 3 ) was then used to segment bone from soft tissue for all distal femur images. Micro-CT analysis was done in a completely blinded manner with all mice assigned to coded sample numbers.
Histomorphometry. Right tibia from 8-week-old mice were subjected to bone histomorphometric analysis. The mice were injected with 20 mg kg À 1 body weight of calcein and 40 mg kg À 1 body weight of demeclocycline on 7 and 2 days before necropsy, respectively. The tibia was dissected and fixed in 70% ethanol for 3 days. Fixed bones were dehydrated in graded ethanol, then infiltrated and embedded in methylmethacrylate without demineralization. Undecalcified 5 mm and 10 mm thick longitudinal sections were obtained using a microtome (RM2255, Leica Biosystems., IL, USA). The 5 mm sections were stained with Goldner Trichome and at least two nonsecutive sections per sample were examined for measurement of cellular parameters. The 10 mm sections were left unstained for measurement of dynamic parameters, and only double-labels were measured, avoiding nonspecific fluorochrome labelling. A standard dynamic bone histomorphometric analysis of the tibial metaphysis was done using the Osteomeasure analysing system (Osteometrics Inc., Decatur, GA, USA). Measurements were performed in the area of secondary spongiosa, 200 mm below the proximal growth plate. The observer was blinded to the experimental genotype at the time of measurement. The structural, dynamic and cellular parameters were calculated and expressed according to the standardized nomenclature 70 .
Statistics. All experiments were performed at least twice. Data are expressed as means of triplicate biological repeats within a representative experiment plus/minus standard error. Statistical analyses were peformed using an unpaired two-tailed Student's t-test (Microsoft Excel), P-values o0.05 were considered to be significant. Variation between groups was similar in all cases.
Data availability. The RNA-seq data are deposited in GEO under accession code GSE76932. The authors declare that all other data supporting the findings of this study are available within the article and its supplementary information files.
Corrigendum: SIKs control osteocyte responses to parathyroid hormone